#

Key takeaways

• Google introduced two new audio-to-audio models for real-time voice applications.

• One version focuses on scale and cost efficiency.

• The other version targets higher-complexity reasoning.

• The models support visual inputs, background tool calls, and 97 languages.

• Rollout is through the Gemini API, Google AI Studio, Search Live, and selected products.

What Google announced

Google announced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on September 15, 2026. The company describes both as audio-to-audio models for real-time voice applications. The announcement separates the two versions by intended use.

Gemini 3.8 Live is presented as the scale

• and cost-efficiency option. Gemini 3.8 Live Extended Thinking is presented as the higher-complexity reasoning option. That distinction is the main product framing in the source.

What the models are designed to do

The source says the models are built for real-time voice applications. It also says they support visual inputs and background tool calls. Those capabilities suggest a broader multimodal setup, but the source does not go beyond that description.

Google also says the models support 97 languages. That is a company-reported capability. The source does not provide independent verification or performance detail.

Where the models will appear

Google says rollout will happen through the Gemini API, Google AI Studio, Search Live, and selected enterprise and consumer products. The source does not list every product in that rollout. It also does not establish when each rollout step will reach every user.

This matters because availability can differ by channel. The announcement gives a distribution path, not a universal release statement. Readers should treat the rollout details as company-reported.

How to read the two versions

The naming suggests a split between efficiency and deeper reasoning. Gemini 3.8 Live is the lighter option in Google's description. Gemini 3.8 Live Extended Thinking is the more complex option.

That structure may help teams choose a model based on task needs. For simpler voice workflows, the efficiency-focused version may fit better. For more demanding reasoning tasks, the Extended Thinking version may be the better match. This is an interpretation of the product labels, not a claim about measured outcomes.

Operational considerations

The source points to three practical considerations. First, teams should match the model to the task. Second, they should account for tool calls in workflow design. Third, they should plan for multilingual use if language coverage matters.

The announcement does not provide benchmark details in the input beyond saying they are company-reported. It also does not provide independent comparisons. So any evaluation should stay close to the stated product roles and rollout channels.

What is not established in the source

The input does not establish Moroccan availability. It also does not establish local pricing, local partners, or local deployment terms. Those details are outside the supplied source.

The input also does not confirm any country-specific market impact. It only reports Google's announcement and the company's rollout framing. Any broader interpretation should remain conditional.

Morocco relevance

The source reports no Morocco-specific fact. For readers, the global lesson is simple: check the announced rollout channel and product fit before assuming access or suitability.

Bottom line

Google's announcement adds two new options for real-time voice applications. One emphasizes efficiency, while the other emphasizes more complex reasoning. Both include visual inputs, background tool calls, and support for 97 languages.

The key limitation is scope. The source is a company announcement, and it does not establish independent benchmark results or Morocco-specific availability. Readers should treat the release as a product update, not a local deployment confirmation.